Following up on the improvements in CDRIVER-4002 as part of PERF-2203, we are still seeing some lesser contention in mongoc_ts_pool_return. We are seeing between 38 and 136 samples out of 2500 samples waiting on this lock.

This is far better than before the upgrade with CDRIVER-4002, but may be responsible for part of the 14 ms latency we're seeing.

Please loop me in to any investigations here; now that TIG-3599 has been done we can revendor the C++ driver into genny very quickly and easily, so can very tightly iterate with y'all on this.